Capability or Gage R&R Study for Leak Tester

OmarEn

Involved In Discussions
#1
Hello guys,

I have a really hard trouble making an analysis of my leak tester

We put the pipe plugged into the device (leak tester)with the opposite side blocked, then 150 psi air pressure fill the interior of pipe in order to verify any possible leak.
Once this device has completed the test, It displays a leak value which It should not be greater than the leak value specified in the print. If the value is greater, the part has failed because It has a leak.

Logically, this is not a replicable test, because same part never will show same values. We can not control the value, we can only control if the part is going to leak or not by checking all joints where It could be a leak. That is the most important thing here.

The problem is that a normal capability study does not represent anything here since we do not control the distribution of the data and I do not care if I have a CPK 1.33 in this case.

Someone recommended me a Gage R&R study Nested. I have already done that in Minitab but I can not find how that can help this.

Does anybody of you guys know something about how to conduct a kind of capability or Gage R&R Study for this type of situations? I really need some help.

Thanks in advance.
 

Bev D

Heretical Statistician
Staff member
Super Moderator
#2
Hi OmarEn! :bigwave:

Are you trying to perform a capability study on the pipe seal or are you trying to perform a measurement study on the lead tester?

A capability study is quite straightforward...from your post I think you are trying to perform a measurement study so I'll address that first.

Logically, this is not a replicable test, because same part never will show same values.
well, no measurement system ever returns the same values regardless of the thing being measured. That's just physics - and occasionally geometry. The purpose of the MSA is to understand how much your measurement system actually varies in relationship to the variation of the actual parts and the specification range. you are answering the question how much can I trust this system? This is especially critical when doing acceptance testing as you appear to be doing. if the measurement system variation is large enough (either precision or accuracy) and parts are near the tolerance you will get false fails and false rejections. This may also lead to the dreaded "retest to pass" when the system is of questionable believability.

this is really a simple test to perform unless the test is destructive which it doesn't sound like it is. This sounds like a simple non-destructive evaluation (NDE) please correct me if I have this wrong.

you can look at my paper on functional testing which also covers NDE testing to understand how to set up the study. Essentially get 15-30 pipes with various values that span your expected or historical range of variation and test them twice. You can post the data here for additional advice and insight...

We can not control the value, we can only control if the part is going to leak or not by checking all joints where It could be a leak. That is the most important thing here.
Not sure what you mean by this? can you clarify? are you the manufacturer or the Customer of the pipe? or are you a test house?

a capability study and an MSA have a lot of value regardless of your ability to control the 'value'. unless by controlling the value you mean that you cannot deliberately create pipes of varying leak rates? if that is what you mean you don't have to create pipes of various values (and that would be wrong anyway; just select a good sample.

The problem is that a normal capability study does not represent anything here since we do not control the distribution of the data and I do not care if I have a CPK 1.33 in this case.
Again not sure what you mean by this? can you clarify?

Someone recommended me a Gage R&R study Nested. I have already done that in Minitab but I can not find how that can help this.
It doesn't sound like a nested study is required but It might be helpful if we could see the data. can you post the data? (in EXCEL for those of us who don't minitab)
 

OmarEn

Involved In Discussions
#3
Hi Bev D,

Thanks for looking and answer my question.

I am the manufacturer of that pipe and we are trying to conduct an MSA study (I do not know what type of). We manufacture brake pipes for GM. Those pipes have been torqued at one end and crimped in several places to join hose and pipe. If we leak test only the pipe by itself, there will not be a leak at all. But when all those joints have been made, the line can leak at one or more of those joints. We have a specification for leak test (0.700 ccm max).
That's why I said we can not control the value of leak. we'd rather control that part simply not leak (at least not more than 0.700 ccm).

Please find attached the excel worksheet with the data.
they are 30 parts and its leak value. All of these 30 samples have been taken randomly. All of them are good.

Thanks in advance for taking your time to answer my question.

Best Regards
Omar Ríos.
 

Attachments

Bev D

Heretical Statistician
Staff member
Super Moderator
#4
i don't understand the structure of the test that this data represents. it looks to be a series of single measurements of 30 parts. can you confirm? if it is this is not a nested design (or a crossed design) and cannot be used to evaluate the measurement variation.

do you have any questions on my response above?
 

Miner

Forum Moderator
Staff member
Super Moderator
#5
i don't understand the structure of the test that this data represents. it looks to be a series of single measurements of 30 parts. can you confirm? if it is this is not a nested design (or a crossed design) and cannot be used to evaluate the measurement variation.
Bev,
Per the OP's comment below, this appears to just be 30 randomly selected samples (not in time sequence) and with no replication.

From the ugly histogram, I suspect we have a mixture. The pipe probably has multiple joints, each of which has a different opportunity for a leak as well as a different magnitude of leak. Unfortunately, the OP's test method (presumably pressure decay) does not allow identification of the leak location. Otherwise, I would recommend a multi-vary plot.

Omar,
Do you have access to a He leak detector that would allow you to identify the leak locations on these parts?

Please find attached the excel worksheet with the data.
they are 30 parts and its leak value. All of these 30 samples have been taken randomly. All of them are good.
 

Omar.

Starting to get Involved
#6
Hello Miner & Bev D,

Miner,
Yes, they are just 30 randomly selected samples (not in time sequence) and with no replication. and you are right when you said:
"The pipe probably has multiple joints, each of which has a different opportunity for a leak as well as a different magnitude of leak. Unfortunately, the OP's test method (presumably pressure decay) does not allow identification of the leak location. Otherwise, I would recommend a multi-vary plot."

and No, we can not know the location of the leak.
could you recommend me a type of study for this situation? how can we evaluate the leak tester?

thanks in advance for your time.
 

Miner

Forum Moderator
Staff member
Super Moderator
#7
If you still have the 30 parts and can link the leak rates to the specific parts, I would test those same parts two more times. This will provide the data necessary to perform a Repeatability study. As Bev stated, any measurement device will provide varying results when measuring the same parts. The only exception to this is when the measurement device has inadequate resolution to see the variation.
 

Omar.

Starting to get Involved
#8
Hello guys,

I have already done the type 1 study.
but I can not understand the results (attached).. What are CG and CGK and how are they calculated? what is T value?

thanks in advance.
 

Attachments


Top